Spaces:
Runtime error
Runtime error
Update app.py
Browse filesReverted to Slovak after runtime error
app.py
CHANGED
@@ -14,8 +14,8 @@ asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base",
|
|
14 |
# load text-to-speech checkpoint and speaker embeddings
|
15 |
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
|
16 |
|
17 |
-
|
18 |
-
model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
|
19 |
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
|
20 |
|
21 |
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
|
@@ -23,7 +23,7 @@ speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze
|
|
23 |
|
24 |
|
25 |
def translate(audio):
|
26 |
-
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "
|
27 |
return outputs["text"]
|
28 |
|
29 |
|
@@ -42,8 +42,8 @@ def speech_to_speech_translation(audio):
|
|
42 |
|
43 |
title = "Cascaded STST"
|
44 |
description = """
|
45 |
-
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in
|
46 |
-
|
47 |
|
48 |
![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
|
49 |
"""
|
|
|
14 |
# load text-to-speech checkpoint and speaker embeddings
|
15 |
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
|
16 |
|
17 |
+
model = SpeechT5ForTextToSpeech.from_pretrained("aoliveira/speecht5_finetuned_voxpopuli_sk").to(device)
|
18 |
+
#model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
|
19 |
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
|
20 |
|
21 |
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
|
|
|
23 |
|
24 |
|
25 |
def translate(audio):
|
26 |
+
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "sk"})
|
27 |
return outputs["text"]
|
28 |
|
29 |
|
|
|
42 |
|
43 |
title = "Cascaded STST"
|
44 |
description = """
|
45 |
+
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Slovak. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
|
46 |
+
SpeechT5 TTS model fine tuned for the Slovak language for text-to-speech:
|
47 |
|
48 |
![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
|
49 |
"""
|